Performance of a 23 years TOPAZ reanalysis L. Bertino, F. Counillon, J. Xie,, NERSC LOM meeting, Copenhagen, 2 nd -4 th June 2015
Outline Presentation of the TOPAZ4 system Choice of modeling and assimilation tools The 23-years physical reanalysis Good health of an Ensemble Kalman Filter Ocean variables Sea ice variables Plans for TOPAZ5 /perspectives
The HYCOM model at NERSC 3D numerical ocean model Hybrid Coordinate Ocean model, HYCOM (U. Miami) Horizontal resolution 12 km Conformal mapping: uniform Hybrid vertical coordinate Isopycnal in the interior Z-coordinate at the surface No sigma layers TOPAZ4 uses 28 layers Hybrid coordinates in the Arctic Sharp pycnocline Less spurious diapycnal mixing (critical at high model resolution)
Local HYCOM settings 4 th Order scheme, momentum advection Used in near real time, but not in this reanalysis Sea ice coupled with HYCOM CICE V3, NERSC thermo No coupler, same code Still using sigma-0 Rivers using ERA-Interim + TRIP No SSS relaxation if Delta S > 0.5 psu
Ensemble Kalman filtering Forecast Analysis Member1 1 2 Member2 Member99 1. Initial uncertainty 2. Model uncertainty (winds, surf. Tem) 3. Measurement uncertainty Observations 3 Member100
Why dynamic Data Assimilation in the Arctic? Example of ice-salinity correlations in the Barents Sea ice formation/ melting warm+ salty AW flux Sakov et al., the TOPAZ4 system, OS 2012 Also see Lisæter et al. Oc. Dyn. 2003
Comparison of dynamical to static / climatological covariances Mobile ice edge = mobile covariances
The TOPAZ system at a glance
Assimilation DEnKF, asynchronous 100 members Local analysis (~90 km radius) Ensemble inflation by 1% Observations: Sea Level Anomalies (CLS) SST (NOAA, then UK Met) Sea Ice Concentr. (OSI-SAF) Sea ice drift (CERSAT) T/S profiles (Coriolis, IPY) 400.000 observations per week ~100 in each local radius SRF: local spread reduction factor
The TOPAZ system again Exploited operationally at MET Norway Since 2008 Ecosystem coupled online 20 years reanalysis at NERSC Took 2 years to produce 3-years ecosystem reanalysis Assimilation of both physical and ocean colour data MyOcean/Copernicus Arctic MFC Free distribution of data RT Data used by ECMWF wave forecast model Surface currents Ice thickness forecast for 11 th May 2015
Data assimilation statistics SLA Stable / decreasing errors Stable ensemble spread
In situ TEM innovation statistics IPY ITP profiles +1-1 Smaller RMS errors Smaller bias Depths 300-800 m
In situ SAL innovation statistics IPY ITP profiles +0.1-0.1 Smaller RMS errors Smaller bias Depths 300-800 m
Arctic-wide sea level change Low amplitude of seasonal signal Same trend: 2mm/yr Same performance wrt tide gauges No improvements, no degradation
Arctic SST trend Trend: 0.025 K/yr Improved by assimilation (of SST or of sea ice?)
Independent data: surface drifters
Current velocities near surface
Validation of 1993-2009 reanalyses Validation of 1993-2009 reanalyses, focus on vol & heat fluxes, hydrography in the Nordic Seas Global / Arctic MFC NEMO / TOPAZ Monthly means,both free runs and assimilated runs Mean, std, seasonal cycle and trends Lien V., S. Hjøllo, M. Skogen, H. Wehde, E. Svendsen, G. Garric, M. Chevallier, F. Counillon, L. Bertino (in progress)
NEMO Free: Slightly higher salinity, temperature and speed than in assimilated run TOPAZ free: More saline AW core than in assimilated run, but AW depth similar. Too weak currents.
NEMO assim: Realistic hydrography: AW core at Shetland shelf slope; sloping T and S surfaces; AW above ~500 m. Too weak currents TOPAZ assim: Realistic hydrography: AW core at Shetland shelf slope; sloping T and S surfaces; AW above ~500 m.
Fram Strait Water masses
Free run shows a slower trend. Corrected by assimilation (as expected) Icea area anomalies
Ice drift in the model Example 3-days end of March 2013 TOPAZ OSI-SAF
Ice drift (km/d) Ice drift seasonality shortcoming of the EVP rheology 12 10 8 6 4 The seasonal cycle is off phase: buoy Aligned with the wind intensity Ctrl but not sensitive to the cycle of REA ice thickness LM1 2 Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec Month
Ice thickness validation Too thin ice in reanalysis (degradation) Corrected in next V5 system IceSAT indicates smaller seasonal cycle of thickness
Summary performance The Good: Constraint of ice edge within +/- 50km Processes related to presence of ice (mixing, blooms) Most input data respected simultaneously Useful for planning field experiments The Bad: Heavy computational burden Not yet eddy-resolving (planned for 2017) Insufficient advection of Atlantic Water to Arctic Sea ice too thin The Ugly: The sea ice model needs a new rheology to improve the drift Absence of sea ice biogeochemistry model
Evolution until 2018
Next steps TOPAZ5 (2018) Wave-induced mixing in KPP Hourly output in real-time / daily in reanalysis 1 post-doc position soon opened Double horizontal resolution (6 km) Double vertical resolution (50 z-rho layers) Sigma-2* Nesting in global NEMO model Biological model ECOSMO
Increased horizontal resolution Blue here V1: TOPAZ4 (12 km) V4: TOPAZ5 (6 km)
Even further steps Sea ice model in (horizontal) Lagrangian coordinates Consistent with solid mechanics (elastic-brittle rheology) nextsim model (Rampal and Bouillon, OM 2015) Coupling through ESMF. Wish list for HYCOM developments: I/O to NetCDF (r/w access water columns) would make assimilation code much simpler. Better cold halocline representation
A new generation of sea ice models First steps Barents Sea regime Kara Sea regime MIZ regime? Vorticity